powerskills-tool

PowerSkills

PowerSkills is the skill layer on top of PowerMCP. PowerMCP exposes raw software tools. PowerSkills tells an agent which tool to reach for first, what to verify before escalating to advanced studies, and what mitigation playbooks to follow when a study finds a real operating problem.

The skills are packaged as two Agent Skills plugins inside a Claude Code marketplace, so you can install them in one step (see Install).

| Plugin | What it is | Skills | | --- | --- | --- | | powerskills-tool | Progressive-disclosure workflows for power-system software | 10 | | powerskills-engineering | Senior-engineer mitigation playbooks | 10 |

How this repo differs from PowerMCP

1. Progressive disclosure of software tools

• Start with the lowest-risk actions first: open or load the case, inspect the model, solve the base case, and only then move to advanced studies.
• Avoid jumping directly into contingency analysis, dynamics, matrix extraction, or investment optimization before the base case is credible.

2. Senior engineer mitigation skills

• Add issue-driven playbooks for voltage violations, thermal overloads, contingency findings, dynamic stability problems, planning infeasibilities, convergence failures, fault-duty problems, frequency response, interconnection impacts, and DER hosting capacity.
• Move beyond violation reporting into corrective action, tradeoff discussion, and validation steps.

Every powerskills-tool skill ends with an Escalation triggers table that maps a concrete observation (e.g. a bus below 0.95 pu, a branch above 100% loading, a binding N-1) to the matching powerskills-engineering playbook — so the handoff is driven by numbers, not vibes.

Available software skills (powerskills-tool)

These skills mirror the current software list in PowerMCP.

| Skill | Scope | First tools to expose | Advanced tools to expose later | | --- | --- | --- | --- | | ANDES | Dynamic security and small-signal studies | run_power_flow, get_system_info | run_eigenvalue_analysis, run_time_domain_simulation | | Egret | Market and operations optimization | solve_dc_opf | solve_ac_opf, solve_unit_commitment_problem | | LTSpice | Circuit simulation workflow | create_simulation_session, run_simulation | list_available_traces, plot_specific_traces, GUI viewing | | OpenDSS | Distribution feeder studies | compile_and_solve, get_total_power, get_bus_voltages | set_load_multiplier, run_daily_energy_meter, get_harmonic_results | | PSLF | Transmission power flow and contingencies | open_case, solve_case, violation checks | topology or device edits, run_contingency_analysis | | PSSE | PSS/E base-case and API-guided studies | open_case, solve_case | lookup_psspy_command, search_psspy_commands, run_psspy_command | | PowerWorld | Steady-state analysis and sensitivities | open_case, run_powerflow, result queries | contingencies, parameter changes, PTDF or LODF or Jacobian tools | | PyPSA | Planning, OPF, and expansion studies | load_network or create_network, network inspection | optimize_network, optimize_investment, import or export flows | | pandapower | AC analysis and screening studies | load_network or create_empty_network, get_network_info, run_power_flow | run_contingency_analysis | | surge | Transmission analysis, sensitivities, OPF, contingency, ATC, dispatch | load_builtin_case or load_network, get_network_info, run_ac_power_flow | compute_ptdf / compute_lodf, run_dc_opf / run_scopf, run_n1_branch_contingency, compute_nerc_atc, run_scuc |

Available mitigation skills (powerskills-engineering)

| Skill | Use when | Typical actions | | --- | --- | --- | | voltage-violation-mitigation | Low or high voltage, weak reactive support, tap exhaustion | AVR or Volt-VAR checks, shunts, taps, redispatch, transfer reduction | | thermal-overload-mitigation | Line or transformer overloads | redispatch, topology switching, phase-shifter control, reinforcement screening | | contingency-mitigation | N-1 or N-2 violations or weak corrective-action plans | pre-contingency fixes, corrective switching, RAS screening, long-term upgrades | | dynamic-stability-mitigation | Poor damping, transient instability, slow voltage recovery | model checks, dispatch relief, AVR or PSS or governor tuning, dynamic VAR support | | operations-planning-mitigation | OPF or UC infeasibility, high curtailment, congestion, reserve shortage | data cleanup, simpler screening solves, flexibility additions, constraint review | | convergence-failure-mitigation | Power flow diverges or fails to solve in any tool | data checks, island/slack review, staged control relaxation, stress reduction | | short-circuit-mitigation | Fault duty above breaker ratings, rising fault levels | study verification, bus splitting, series reactors, breaker upgrades | | frequency-response-mitigation | Low inertia, poor nadir or RoCoF, weak primary response, UFLS risk | governor headroom, fast frequency response, inertia additions, droop/deadband fixes | | interconnection-impact-mitigation | New generator, storage, or large-load interconnection screening | N-1 screens at the POI, SCR/weak-grid checks, reactive requirements, upgrade sizing | | der-hosting-capacity-mitigation | DER-driven voltage rise, reverse flow, protection desensitization | inverter Volt-VAR/export limits, regulator settings, protection re-checks, reinforcement |

Install

A skill is just a folder with a SKILL.md, so the same content works across agents. Pick the path for your tool.

Claude Code (recommended)

Add this repository as a plugin marketplace, then install one or both plugins:

/plugin marketplace add Power-Agent/PowerSkills
/plugin install powerskills-tool@powerskills
/plugin install powerskills-engineering@powerskills

To work from a local clone instead of GitHub, run /plugin marketplace add ./ from the repository root. Manage everything later with /plugin.

Codex CLI

Codex discovers skills from .agents/skills/ (scanned from your working directory up to the repo root) and from ~/.codex/skills/ (global). Copy or symlink the skill folders you want — no conversion needed:

# project-scoped, e.g. the tool skills for the repo you're working in
mkdir -p .agents/skills
cp -R /path/to/PowerSkills/powerskills-tool/skills/* .agents/skills/

# or install globally for every project
mkdir -p ~/.codex/skills
cp -R /path/to/PowerSkills/powerskills-tool/skills/* ~/.codex/skills/
cp -R /path/to/PowerSkills/powerskills-engineering/skills/* ~/.codex/skills/

Claude Desktop / claude.ai

1. Enable Settings → Capabilities → Code execution (skills run in the code-execution sandbox). 2. Open Settings → Capabilities → Skills → Upload skill. 3. Upload a ZIP of a single skill folder, with the skill folder at the root of the ZIP. Skills are uploaded one at a time. For example:

cd powerskills-tool/skills
zip -r pandapower.zip pandapower

Then toggle the skill on. Repeat for each skill you want.

Repository layout

PowerSkills/
├── .claude-plugin/
│   └── marketplace.json            # makes the repo an installable marketplace
├── powerskills-tool/               # plugin: software workflow skills
│   ├── .claude-plugin/plugin.json
│   ├── README.md
│   └── skills/
│       ├── andes/ … surge/         # 10 tool skills (pandapower, pypsa, surge bundle scripts + cases)
├── powerskills-engineering/        # plugin: mitigation playbooks
│   ├── .claude-plugin/plugin.json
│   ├── README.md
│   └── skills/
│       └── voltage-violation-mitigation/ … der-hosting-capacity-mitigation/  # 10 mitigation skills
└── skill-creator/                  # authoring/validation tooling (not part of either plugin)

Contribution rules

Contributors should use Skill Creator when adding or updating skills.

• Keep SKILL.md concise and procedural.
• For software skills, expose tools in this order: load or open → inspect → solve → modify → advanced studies, and end with an Escalation triggers table.
• For mitigation skills, start from an observed problem and list corrective actions in the order a senior engineer would actually try them.
• Reference bundled assets with skill-relative paths (scripts/..., references/...) so each skill stays portable when installed standalone.
• Prefer contributions in two buckets:
• better progressive disclosure for an existing PowerMCP software surface (powerskills-tool)
• better mitigation playbooks that use one or more of those software tools (powerskills-engineering)
• Validate before opening a PR:

  python skill-creator/scripts/quick_validate.py powerskills-tool/skills/<skill>
  python skill-creator/scripts/quick_validate.py powerskills-engineering/skills/<skill>

Related projects

• PowerMCP - MCP servers for power-system software
• Agent Skills - open skill format and ecosystem

Core Team

• Qian Zhang, Matheus Duarte, Muhy Eddin Za’ter, Drew Gray

License

MIT License